Categorical Approximability for Metric Spaces

• Categorical Approximability for Metric Spaces is a framework that unifies approximate combinatorial, algebraic, and geometric structures with metric data using enriched categorical techniques.
• The embedding theorem demonstrates that approximate categorical structures can inject into genuine metrized categories under conditions like boundedness, separation, and graph transitivity.
• This framework generalizes classical concepts of finite nets and approximate injectivity, with applications spanning Banach spaces, symplectic topology, and persistence module theory.

Categorical approximability for metric spaces refers to a spectrum of frameworks unifying approximate combinatorial, algebraic, and geometric structures with metric data through categorical and enrichment-theoretic machinery. It generalizes classical finiteness and compactness notions and generates new perspectives on approximation, embedding, and injectivity in functional, geometric, and applied mathematical domains.

1. Foundations: Definitions and Structures

The core ingredient is an enriched categorical structure on metric spaces or families of objects equipped with "approximate composition" operations. A metrized category $(\Ob(\mathcal C),\Hom,\circ,\delta)$ consists of:

• Objects $x \in \Ob(\mathcal C)$.
• Hom-sets $\Hom(x,y)$ carrying a metric $\delta_{x,y}: \Hom(x,y)\times\Hom(x,y)\to[0,\infty)$.
• Associative and unital composition $\circ$.
• Metric compatibility: for any $f,f':x\to y$ and $g,g':y\to z$,

$$\delta_{x,z}(g\circ f,\, g'\circ f')\le \delta_{x,y}(f,f') + \delta_{y,z}(g,g').$$

An approximate categorical structure (AC structure) is a triple $(X, \{A(x,y)\}, d)$, where $A(x,y)$ is a set of arrows, and $d(f, g, h)\geq 0$ measures how well $h$ approximates $g\circ f$. The function $d$ generalizes 2-metrics and must satisfy:

• Identity: $d(f,1_y,f) = d(1_x, f, f) = 0$.
• Left associativity (tetrahedral inequality), and
• Right associativity, which induce a collection of triangle-type estimates and pseudometric structures on hom-sets.

These axioms guarantee the "shape" of composition even when associativity or exact closure properties are only approximately satisfied. Such formulations appear in (Aliouche et al., 2015).

A more general framework involves categories enriched over metric spaces ("Met-enriched categories"). Here, hom-sets themselves are metric spaces, and composition is non-expanding. Approximate injectivity (see §4) then provides a further notion of approximation through morphism extension, vital in analysis and operator theory (Adámek et al., 2020).

2. Embedding and Reflectivity Theorems

A principal result for AC structures is the embedding theorem of Aliouche–Simpson (Aliouche et al., 2015):

Let $(X, A, d)$ be an AC structure satisfying boundedness, separation, absolute transitivity, and a non-degeneracy ("graph transitivity"). Then there exists a genuine metrized category $(\mathcal C,\delta)$ with $\Ob(\mathcal C)=X$ such that $> d(f,g,h) = \delta(g\circ f,\,h) >$ for all $f\in A(x,y),\;g\in A(y,z),\;h\in A(x,z)$. The original arrows embed injectively into the morphisms of $\mathcal C$.

The proof constructs a maximal pseudometric on the free category, then identifies a canonical quotient yielding metrized coherence. This guarantees that "almost" categorical or compositional data embeds within a stricter, metric-compatible categorical framework.

More generally, for any class of morphisms $\mathcal M$ satisfying mild smallness (a form of approximate presentability/compactness), the approximate-injectivity class is weakly reflective in Met-enriched categories. Every object admits a canonical morphism to its approximate reflection, constructed via transfinite chains of $\varepsilon$-pushouts (Adámek et al., 2020). This provides an analog of compactification or injective hulls in the metric-enriched categorical world.

3. Notions of Approximability and Metric Nets

Categorical approximability is designed to generalize total boundedness and finite $\varepsilon$-net constructions. For a (pseudo-)metric space $(X, d)$ and a triangulated category $\mathcal C$ on objects $Y\supset X$, $X$ is categorically $\varepsilon$–approximable in $\mathcal C$ if there is a finitely generated full triangulated subcategory such that every point of $X$ lies within $\varepsilon$ (in the metric) of an object in the subcategory.

In persistence module theory, triangulated persistence categories (TPCs) add an interleaving metric $d_\mathrm{int}$, and $\varepsilon$-approximability asks for finite sets such that every object is within $d_\mathrm{int}<\varepsilon$ of a summand or cone generated by the net. This weakens strict compactness: a space can be categorically approximable even when it admits no finite $\varepsilon$-net in the classical sense (Ambrosioni et al., 18 Jan 2026).

The classical notion, by contrast, requires:

• For any $\varepsilon>0$, a finite $\varepsilon$-net $N\subseteq X$ such that every $x\in X$ has $d(x,N)<\varepsilon$.

This distinction is essential in applications to spaces of Lagrangian submanifolds, where metric completions can be non-compact, but spectral or persistence invariants allow for categorical nets with strong approximation properties.

4. Approximate Injectivity and Reflections

Approximate injectivity enables the extension of morphisms up to arbitrarily small errors. In a Met-enriched category $\mathcal K$, an object $K$ is approximately injective with respect to a class of morphisms $\mathcal M$ if every morphism $g:X\to K$ can be approximated, up to prescribed tolerance, by composition with a morphism $u:Y\to K$ whenever $h:X\to Y$ is in $\mathcal M$: $$d_{X,K}(u\circ h,\,g)\le \varepsilon\qquad\text{for all }\varepsilon>0.$$ The class of approximately injective objects always forms a weakly reflective subcategory under mild set-theoretic smallness conditions (Adámek et al., 2020). This leads to canonical "approximate reflections", analogs of injective hulls or universal completions, as exhibited by the construction of the Gurarii space in Banach space theory.

Reflectivity is realized via a transfinite process involving $\varepsilon$-pushouts along the class $\mathcal M$.

5. Examples and Applications

Setting	Notion Realized	Comments
Metric spaces	AC enriched/metric-enriched categories	Lawvere metric enrichment, 2-metrics (Aliouche et al., 2015)
Banach spaces	Approximate injectivity	Gurarii space as weak reflection (Adámek et al., 2020)
Lagrangian submanifolds	Categorical/TPC approximability	TPC-approximable but not precompact (Ambrosioni et al., 18 Jan 2026)
Piecewise linear paths	Area-minimizing composition distance	Category of paths, 2-metric extension

In symplectic topology, certain spaces of Lagrangians (e.g., closed exact in $D^*N$ or equators in $S^2$) are categorically approximable in triangulated persistence categories, but not totally bounded. Approximability is detected via explicit finite families of cotangent fibers or geodesics, with all Lagrangians lying in iterated cones of these approximants in the ambient category, up to prescribed spectral or persistence metric error (Ambrosioni et al., 18 Jan 2026). This has implications for invariants such as the Gromov width and quantitative spectral rigidity.

In classical approximation and clustering, the concept of finite nets and their algorithmic construction (as in metrics of finite doubling dimension) feed into both metric approximability (e.g., via PTAS for $k$-median in doubling metrics (Cohen-Addad et al., 2018)) and categorical approximability via finite bases for persistence categories.

6. Structural Unification and Further Directions

Categorical approximability unifies:

• Metric enrichment of categories (Lawvere, 2-metrics, area-bounded path categories).
• Generalizations of nets, total boundedness, and compactness.
• Weak reflection and injectivity in the metric-enriched context.
• Geometric, algebraic, and analytic applications, from Banach space theory to symplectic topology.

Key open directions include:

• Extensions to higher $n$-metrics and $n$-categorical enrichment (Aliouche et al., 2015).
• Refinement of approximation constants and construction algorithms for nets in spaces with additional structure (doubling dimension, topological, or geometric constraints).
• Broader applications in quantitative concurrency, shape analysis, and enriched homotopy theory.
• Development of fixed-point theory and further rigidity theorems in the presence of categorical approximation data.

The embedding and reflectivity theorems guarantee that approximate composition or extension data in a metric-enriched or persistence-theoretic setting can always be realized inside a genuine metrized categorical framework, providing foundational infrastructure for advanced approximation paradigms in modern mathematics (Aliouche et al., 2015, Adámek et al., 2020, Ambrosioni et al., 18 Jan 2026).